A research team at Seoul National University Bundang Hospital (SNUBH) has developed an artificial intelligence (AI) model that can diagnose sleep apnea by analyzing head and neck X-ray images (cephalograms).

Sleep apnea is a sleep disorder in which breathing is momentarily or intermittently disrupted during sleep. If left untreated, it can result in poor sleep quality, leading to chronic fatigue and drowsiness that can impact daily life. Moreover, it can significantly increase the risk of cardiovascular disease, including high blood pressure, heart attack, and stroke.

Currently, if sleep apnea is suspected, hospitals utilize a screening test to confirm symptoms. Based on the screening test results, additional diagnostic polysomnograms may be conducted to provide a further diagnosis.

Although several screening tests have been developed, they have their limitations. For instance, some have low accuracy rates, while others are not suitable for use in multi-person settings.

In response to these limitations, the research team, led by Professors Jeong Han-gil and Kim Tack-eun of the Department of Neurosurgery, along with Yun Chang-ho of the Department of Neurology, developed a deep learning-based AI model. The model is capable of predicting sleep apnea by analyzing head and neck X-ray images.

Professors Lee Seung-hoon at Korea University Ansan Hospital and Robert Joseph Thomas at Beth Israel Deaconess Medical Center also participated in the study.

The AI model analyzes X-ray images of the patients' heads and necks, with a focus on the upper airway, specifically the tongue and its surrounding structures, which are highly associated with sleep apnea. It can distinguish minute differences that are difficult to identify with the naked eye and classify the presence of sleep apnea accordingly.

The research team created the algorithm by utilizing AI training and validation, which involved using head and neck X-ray image data of 5,591 patients who visited SNUBH. They evaluated the algorithm's performance through internal and external testing procedures.

The results of the testing revealed that the AI model is highly accurate, with an area under the receiver operating characteristics (AUROC) curve of 0.82.

AUROC is an indicator that evaluates the performance of an AI model, and if the figure is closer to one, the better the performance.

The research team highlighted that the head and neck X-ray imaging test required for the diagnosis of sleep apnea has the advantage of being a relatively simple and inexpensive procedure. This finding is significant as it can contribute significantly to improving the diagnosis and treatment rate of sleep apnea, which is crucial for early treatment.

"The prevalence of sleep apnea is estimated to be around 1 billion adults aged 30-69 worldwide, and the number is growing," Professor Yun said. "Early detection and treatment of sleep apnea can prevent further worsening of symptoms and improve quality of life."

Professor Jeong also said, "The team developed an AI model that can screen for sleep apnea using only head and neck X-ray images without other clinical predictors."

With its accuracy and affordability, the team expects the model to play a major role in the early diagnosis and treatment of sleep apnea, Jeong added.

The study results were published in the Journal of Clinical Sleep Medicine.